Reinhard Schuster
CALIBRATION OF THE LH SYSTEMS ADS40 AIRBORNE DIGITAL SENSOR
Reinhard SCHUSTER*, Bernhard BRAUNECKER**
*Deutsches Zentrum für Luft- und Raumfahrt (DLR), Germany
Institute of Space Sensor Technology and Planetary Exploration
reinhard.schuster@dlr.de
**Leica Geosystems AG, Switzerland
bernhard.braunecker@leica-geosystems.com
Working Group IC-4
KEY WORDS: Photogrammetry, Camera, CCD, Sensors, Calibration, Radiometric Qualities
ABSTRACT
High performance photogrammetric cameras require high resolution geometric, spectral and radiometric calibration.
Instruments to meet the stringent specifications exist at LH Systems and DLR. The underlying principles of operation
are discussed with respect to the new LH Systems ADS40 airborne digital sensor, and the results of the first cross
measurement are reported.
1 INTRODUCTION
The rapid progress of electronic sensors will undoubtedly lead to the substitution of film-based photogrammetric
cameras by digital cameras. However, both analogue and digital cameras will be used in parallel by the
photogrammetric community for at least one decade. With the advent of the ADS40, a joint project of LH Systems and
Deutsches Zentrum für Luft- und Raumfahrt (German Aerospace Centre – DLR), the issues of sensor calibration must
be addressed. This paper reviews the calibration equipment currently in use at both locations. While the DLR equipment
is favoured owing to its great flexibility and the possibility of modifying geometric, radiometric and spectral
measurements in the development phase, the LH Systems equipment is optimised to industrial needs, especially through
fast and automatic measurement modes.
2 THE CALIBRATION-FACILITY AT DLR - EXAMPLES OF CALIBRATION RESULTS
The calibration facility located in the DLR Institute of Space Sensor Technology was originally developed for
calibrating spaceborne sensors (Schuster, 1994) and was later extended to the infrared spectrum. The optical scheme of
the calibration set up is shown in figure 1.
cal-source
halogen, hg
or Xe
filter
lightmixing-rod
slit-spot
stepper
vis-collimator
blackbody
rotation-tilt-table
Vis-Sensor
theodolite
MWIR
lightsource
shutter
monochromator
optics
LWIR
x-z-stage
autocollimationtelescope
R-source
IR-collimator
Figure 1. Calibration facility at DLR (optical scheme)
288
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
Reinhard Schuster
The DLR facility (figure 2) features:
• Clean room class 10000
• Lamps: halogen (150W), xenon (150W), mercury (200W)
• Collimator f=1200 mm, D=150 mm, apochromatic corrected 390-1013 nm, with autocollimation device, adjusting
slit or spot: aperture ±25 mm, accuracy ±1 µm, resolution 0.1 µm
• Spectral resolution with grid-monochromator (350-800 nm, 700-1500 nm); ∆λ = 2...20 nm
• Two axis nodal bench azimuth axis 350°, elevation axis ±50°, accuracy ±2.5 arcsec, resolution 0.5 arcsec, load 50
kg
• PC-control of lamps, shutter, monochromator-drive, optical sensors, adjusting aperture and camera nodal bench
• Special software (MS Windows) for geometric and radiometric calibration tasks.
Figure 2. View of the calibration laboratory at DLR-Berlin
The main tasks for geometric calibration of a digital sensor are:
• Highly accurate measurement of the internal orientation of any active pixel for determination of the image coordinates
• Measurements of the system MTF for any pixel of the sensor.
The first requirement is met by single pixel illumination along the CCD-line with a pinhole spot from the collimator
focus. The direction of the illuminated pixel to the collimator axis is determined by the angles α (angle in line direction)
and β (angle perpendicular to the line direction). These two angles can be exchanged in the case of horizontal mounting
of the camera. For the transformation from the spatially fixed co-ordinate system (x,y,z) to the image co-ordinate
system (x´,y´,z´) the simple transformation x' = f ⋅ tanα / cosβ ; y' = f ⋅ tan β ; z' = z (figure 3) must be applied (Ohlhof
and Kornus, 1994). As example the result of calibration from the nadir-line of the first engineering model of the ADS40
(figure 4) is shown in figure 5 before and after transformation.
Figure 3. Transformation from (x,y,z) to (x′,y′,z′)
Figure 4. The engineering model of the ADS40 during
the calibration procedure
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
289
Reinhard Schuster
0.2
60
40
20
Alpha 0
i
Alppix 0
i
20
40
0.2
60
30
20
10
0
10
20
0
30
2000
4000
6000
8000
4
1 10
Betpix
i
Beta
i
Figure 5. Result of geometric calibration in degree (1) before and
in units of pixel-size (2) after transformation
The second requirement is met by measurements of the Point Spread Function (PSF), from which the Modulation
Transfer Function (MTF) can be calculated by a Fast Fourier Transform (FFT). It is useful to register at the same pixel
the PSF and geometry. Because the PSF is a two-dimensional function, we measure in the two main directions
corresponding to the flight direction and perpendicular in the line direction. From these measurements the MTF can be
calculated at an arbitrary number of points along the line and used as input for the image post processing. Examples for
PSF/MTF measurements with the engineering model of the ADS40 are shown in figure 6. The MTF curves characterise
the optical performance of the digital camera system, which is mainly determined by the pixel size and the optics. The
resolution always seems to be better in the flight direction than in the line-direction.
flight-direction
swath direction
x04012n1
MTF
y04012n1
1.0
x04012t1
y04012t1
%
100
1.0
80
60
-10°
0.5
0.5
40
20
0.0
0.0
0
10
20
30
40
50
0
0
10
20
30
40
50
µm
0
50
100
µm
x05938n1
y05938n1
1.0
150
mm-1
x05938t1
y05938t1
%
100
1.0
80
60
0°
0.5
0.5
40
20
0.0
0.0
0
10
20
30
40
50
0
0
10
20
30
40
50
0
50
100
µm
µm
x07479n1
y07479n1
1.0
150
mm-1
x07479t1
y07479t1
%
100
1.0
80
60
+10°
0.5
0.5
40
20
0.0
0.0
0
10
20
30
40
50
µm
0
0
10
20
30
40
50
µm
0
50
100
150
mm-1
Figure 6. Examples of PSF/MTF measurements (nadir line of engineering model of the ADS40)
The use of modern digital sensors together with a flexible electronic read-out channel also leads to new possibilities for
the radiometry. If the total field of view (FoV) of a line camera is illuminated homogeneously we get a line-signal
290
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
Reinhard Schuster
similar to that in figure 7a. The shape of the curve reflects the intensity drop across the FoV, known as the cos4 law, but
the high frequency variation is caused by the variation of responsivity (photo response non-uniformity - PRNU) of the
line-pixels. It seems useful to correct the signals pixel-wise to obtain constant signals with homogenous illumination.
The correction can be done very simply online by subtraction (dark signal) or multiplication (PRNU and limb
shadowing) by the digital part of the electronics. An alternative solution would be off-line correction during the image
processing work. In both cases it is necessary to determine the radiometric correction values in a separate calibration
process. After this correction the line signal must be free from PRNU and limb shadowing (figure 7c).
a
b
pk1a_140 pk1a_180 pk1a_220 pk1a_260 pk1a_300 pk1a_340 pk1a_380 pk1a_420 pk1a_460 pk1a_500 pk1a_540
c
KAF1A1K1
E+3
7.00
tesk1a14 tesk1a18 tesk1a22 tesk1a26 tesk1a30 tesk1a34 tesk1a38 tesk1a42 tesk1a46 tesk1a50 tesk1a54
E+3
7.00
3.50
6.00
6.00
3.00
5.00
5.00
2.50
4.00
3.00
4.00
3.00
2.00
2.00
2.00
1.50
1.00
1.00
0.00
0.00
1.00
0.0
2.0
4.0
6.0
8.0
10.0
12.0
0.0
2.0
4.0
6.0
8.0
10.0
E+3
12.0
0.0
2.0
4.0
6.0
8.0
E+3
10.0
12.0
E+3
Figure 7. Line-signal: a-uncorrected, b-PRNU-correction values, c-corrected
Normally CCD camera systems are sensitive in a relatively wide spectral band. Therefore it is not possible to determine
the spectral behaviour of radiance of an unknown source. Nevertheless the absolute calibration can be done to an
averaged value given by the convolution of the normalised responsivity with the spectral characteristics of the source.
This absolute calibration is a part of the calibration program of the DLR facility. To proceed the following steps are
necessary:
1. Measurements of the irradiance in front of the camera optics with separate detector dependent on wavelength.
2. Measurement of the spectral responsivity of the pixels and normalisation with the spectral behaviour of the
facility.
3. Measurement of an absolute calibrated source (radiometric sphere) covering the total FoV of the camera.
4. Calculation of the absolute radiometric correction values.
The method of absolute calibration was tested with a spaceborne camera system and gave a Noise Equivalent Radiation
(NER) of 30-70 104W/m³sr with a resolution of 7 104W/m³sr. The performance of the PRNU correction was better than
0.2%. Figure 8 shows some sample results.
BDS5_UW
RADI_LB2
E+6
150
6.00
5.00
100
4.00
3.00
50
2.00
1.00
0.00
0
400
500
600
700
800
900
400
500
600
nm
700
800
900
nm
Figure 8. Spectral responsivity of (a) a pixel, (b) radiance of the radiometric W/m³sr
3 PHOTOGRAMMETRIC CALIBRATION FACILITIES AT LH SYSTEMS
The main optomechanical parts of the ADS40 are produced by LH Systems in Heerbrugg, Switzerland. Consequently,
new instruments were developed for dual use at this location: to test lenses during the manufacturing process and to
perform the final calibration of the integrated camera head at system level.
3.1 Geometric Calibration
Geometric calibration of the ADS40 includes the quantitative determination of the image quality and of the registration
geometry. To speed up measurement, both criteria should be measured simultaneously. This is accomplished by a coded
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
291
Reinhard Schuster
vertical goniometer (CVG). The test programs are run automatically and the results directly linked to the LH Systems
databases.
3.2 CVG Hardware
The CVG was developed in 1998 by upgrading an existing vertical goniometer. Its main goal was the electronic testing
of film-based analogue lenses under realistic operational conditions. Later, with the advent of digital cameras, hardware
and software modifications were added to test the lenses together with the digital camera head (Pacey et al., 1999).
Figure 9. Inserting optical systems into the CVG:
a) objective UAGS for testing the optics
b) engineering model of the ADS40 for calibration
Figure 9 shows the installation of test lenses into the CVG, whereby the lens entrance pupil must coincide with the
swivel axis of the goniometer. Inside the swivel arm an auxiliary optics images a small code pattern to infinity. The
ADS40 lens under test refocuses the code pattern on to its digital sensors. Rotating the swivel performs a field scan
along the CCD-line at nadir position. However, to address pixels outside the nadir line, a mirror scanner, moveable in
the x-direction, is mounted on top of the goniometer arm. From the contrast of the transferred image we deduce values
for the image quality; from the actual angles of swivel and x-scanner we gain the registration properties.
The code pattern consists of a series of stochastically arranged black and white bars of width 3 µm pointing in two
orthogonal directions. The code is designed to be sensitive to small lateral displacements to increase the accuracy for
the registration measurement, and to possess sufficient spatial bandwidth for the image quality tests.
3.3 Measurement of the Image Quality
As criteria for the image quality of a pixel the OTF (optical transfer function) across and along the CCD-lines is chosen.
The OTF describes the complex variation of the transferred contrast through the optics with the spatial frequency. To
ensure that the digital sensor lines are at best focus, the OTF must be measured at different depths of focus. This is
easily accomplished by defocusing the small code pattern in the swivel arm.
Figure 10a shows the results of a measurement: The ideal code pattern (red) has contrast one, while the measured code
signals (blue) show smaller contrast values. The OTF is mathematically deduced from both functions. Its amplitude
function MTF (modulation transfer function, upper curve in figure 10b, is already visible in the measured signals in
figure 10a, where code sequences with low frequencies exhibit a larger contrast than those with high frequencies. Note
further that the OTF’s phase function PTF (lower curve in figure 10b) is nearly zero over the full frequency range, thus
indicating a perfect, coma-free test lens. Again, this is nicely visualised in figure 10a, where the shape of all measured
code elements is highly symmetric.
292
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
Reinhard Schuster
3.4 Measurement of the Geometric Registration
The purpose of the registration calibration is to assign to each sensor pixel the angular position of its seen ground pixel
element.
OTF Measurement
1
0.8
0.6
0.4
0.2
0
50
100
150
200
250 300
position (pixel)
350
400
450
500
1
)) /
0.5
0
MTF>0.6:
-0.5
0
MTF>0.3:
MTF>0.1:
50
100
150
200
frequency (lp/mm)
250
300
ν< 22.8 lp/mm
ν< 60.9 lp/mm
ν< 150.7 lp/mm
Figure 10. (a) code pattern before (red) and after (blue) imaging through a test lens
(b) transfer functions MTF (blue) and PTF (green) deduced from (a)
Unlike film cameras, where film shrinking requires the need of artificial fiducial marks, the situation is much easier for
the digital camera. Each CCD-pixels is a fiducial per se. This allows to reduce the calibration task to assign to each
pixel the value of its 'down to ground looking' direction vector a, mostly expressed by two polar angles, relative to a
lens based co-ordinate system.
Owing to the stringent accuracy requirements of 1 arcsec for both angles, special electro-optical means are foreseen in
the CVG to control mechanical motions of swivel and x-scanner in real time. Powerful algorithms, specially adapted to
the stochastic properties of the code structure, allow determination of the actual position of the code at the CCD-lines
with sub-µm resolution.
3.5 Radiometric Calibration
The radiometric calibration is done with a large Ulbricht sphere, located directly beside the CVG. The test bench is still
under construction, but operates very similarly to the DLR equipment. At the time of writing the first test runs with the
engineering model of the ADS40 are complete.
4 COMPARISON OF THE GEOMETRIC CALIBRATION MEASUREMENTS
Photogrammetric cameras must be geometrically calibrated with an accuracy of parts of one pixel. Therefore very
precise angular measurements are necessary to meet the stringent requirements. Using the first working ADS40 model,
the engineering model, geometrical calibration was performed at both calibration facilities of DLR and LH Systems for
the purpose of comparison. The outputs of the geometrical calibration process are two polar angular co-ordinates,
assigned to each measured pixel. The first results indicate that it is sufficient to measure pixels every 2-5 degrees within
the field of view, depending on the length of the CCD-line and on the focal length of the camera. The angular coordinates for pixels in between are interpolated numerically. The geometrical calibration results can be presented in
different forms and units, e.g. in multiples of a pixel. The conversion to conventional units e.g. SI units, is then a simple
multiplication with the pixel size.
In the preceding chapters we outlined the different measurement modes at both locations. At DLR in Berlin the ADS40
is mounted horizontally on a rotation-tilt stage and the direction of the pixel through the optics to the fixed collimator
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
293
Reinhard Schuster
axis is measured by moving the camera around two perpendicular axes. At LH Systems in Switzerland the camera is
fixed in the vertical direction and the illumination device moves within the FoV of the camera. The direct comparison
of the results for the nadir line of the engineering model at different times and places is presented in figure 11. The
measured angles are converted to distortion values, expressed in pixel units. The difference between both curves is less
than one pixel over the whole line of 12,000 pixels. This impressive result indicates the high level of precision and the
high reliability of both devices.
1,5
1
distortion in pixels
0,5
0
0
2000
4000
6000
8000
10000
DLR-stage
-0,5
-1
CVG
-1,5
number of pixel
Figure 11. Comparison of the calibration results at DLR and LH Systems for the nadir line of the engineering model
REFERENCES
Ohlhof, T., Kornus, W., 1994. Geometric calibration of digital three-line CCD cameras. International Archives of
Photogrammetry and Remote Sensing, Vol. 30, Part 1, Como, Italy, pp. 71-81.
Pacey, R.E., Walker, A.S, Scheidt, M., 1999. Calibration of analog and digital airborne sensors at LH Systems.
Proceedings of the 1999 ASPRS Annual Conference, Portland, Oregon, pp. 950-956.
Schuster, R., 1994: Sensor calibration and geometric calibration of a three line stereo camera. International Archives of
Photogrammetry and Remote Sensing, Vol. 30, Part 1, Como, Italy, pp. 265-271.
294
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.